1 / 10

ALERT: Awareness and Localization of Explosives-Related Threats

Prof. Miguel Vélez -Reyes Univ. of Puerto Rico at Mayaguez N. Santiago, V. Manian , UPRM E. Miller, Tufts D. Castañón , W.C. Karl, BU. F3-D: Multimodal Pattern Recognition. ALERT: Awareness and Localization of Explosives-Related Threats. The problem of interest. IR Hyperspectral.

teddy
Télécharger la présentation

ALERT: Awareness and Localization of Explosives-Related Threats

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Prof. Miguel Vélez-Reyes Univ. of Puerto Rico at Mayaguez N. Santiago, V. Manian, UPRM E. Miller, Tufts D. Castañón, W.C. Karl, BU F3-D: Multimodal Pattern Recognition ALERT: Awareness and Localization of Explosives-Related Threats

  2. The problem of interest IR Hyperspectral Pattern Recognition System THz and Microwave Probes X-ray Imaging • We want to • Increase detection rates • Decrease false alarms with high throughput for standoff and portal-based detection systems

  3. Challenges • High dimensionality • Variability introduced by • Changes in atmospheric conditions • Differences in illumination, orientation, etc • Variable unstructured clutter in standoff applications • Mixed signatures (clutter and threat) • Integration of human in the decision process

  4. Proposed Approach • Powerful Methods for Constructing Detectors and Classifiers • Kernel-based methods • Support Vector Machines (SVM) • Adaptive Boosting Techniques • AdaBoost • Dimensionality Reduction • Invariant features • Adaptation • Changing environment • Robust detection of new classes of explosives • Integration with human operator • Decision pre-processors

  5. Our Expertise • Automatic target recognition and optimal sensor management • Classification and detection in high dimensional feature spaces • Hyperspectral image processing • Novel methods for wide range of problems across many application areas in image formation and segmentation from multimodal data • Geometric methods • Probabilistic modeling • Ill-posed inverse problems • Optimization and computation

  6. Spectral Data Pattern Recognition as Detection Aids • Problem of Interest: Automated tools to rapidly process large volumes of data • Tailored to focus attention of human operators when more information is needed • Desired feature: • Classify with great confidence whenever possible • Identify ambiguous cases where additional information is needed (sensor management!) • Results: New theories for classifiers that determine when additional information is needed • Extensions of kernel support vector machines and adaptive boosting • Work in concert with human-in-the-loop • Applied to medical diagnosis Decision Regions with ambiguous class

  7. Positive Matrix Factorization Unsupervised Unmixing: Target Clutter Separation Endmember Determination Endmember Signatures Abundance Estimation Hyperspectral Image Unsupervised Unmixing Abundance Maps

  8. Alternative platforms for hyperspectral image processing • Problem of Interest: Study alternative platforms where hyperspectral algorithms may be mapped efficiently, • Algorithm • Unsupervised unmixing • Platforms • Massively parallel processors – CUDA GPGPUs • Field programmable gate arrays - FPGAs • Features: • Embarrasingly parallel structure • Preliminary Results: Implementation of Image Space Reconstruction Algorithm for abundance estimation on FPGAs and CUDA has resulted in reduction of three orders of magnitude in execution time. • Tune application to platforms.

  9. Novel HSI Spatial/Spectral Processing: Geometric PDE Processing of HSI SEBASS Image • Improve Target Background Contrast • Improve Detection and Classification

  10. Year 1 Work Plan • Initial projects: • Investigated kernel-based methods and adaptive boosting techniques for constructing and updating classifiers • Extension of SVM adaptive classifier developed for biomedical applications • Threats, non-threats and ambiguous objects • Unsupervised unmixing (close collaboration with F2A) • Speed up using hardware implementations • Integration of libraries, a priori information and other data sources • Fusion of multi-sensor classification for portal applications • Initial focus on luggage inspection (Collaboration with F3A) • Maintain close relationship with industrial partners via constant personnel exchange

More Related